You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

210 lines
7.5 KiB

#include <medusa/bits/approximations/RBFFD.hpp>
#include <medusa/bits/approximations/Gaussian.hpp>
#include <medusa/bits/approximations/Operators.hpp>
#include <medusa/bits/approximations/Monomials.hpp>
#include <medusa/bits/approximations/Polyharmonic.hpp>
#include <medusa/bits/approximations/RBFBasis.hpp>
#include <medusa/bits/types/Vec.hpp>
#include <Eigen/Cholesky>
#include <Eigen/LU>
#include "gtest/gtest.h"
namespace mm {
TEST(Approximations, RBFFDGauss2D) {
double h = 0.1123;
double s = 1.5;
Gaussian<double> g(s);
RBFFD<Gaussian<double>, Vec2d> appr(g);
Range<Vec2d> support = {{0, 0}, {0, h}, {h, 0}, {0, -h}, {-h, 0},
{-h, h}, {-h, -h}, {h, -h}, {h, h}};
appr.compute({0.0, 0.0}, support);
double a = -4.*(s*s + h*h/std::pow(std::sinh((h/s)*(h/s)), 2)) / std::pow(s, 4);
double b = 4.*std::exp(3.*(h/s)*(h/s))*h*h / std::pow(-1+std::exp(2*(h/s)*(h/s)), 2)
/ std::pow(s, 4);
Eigen::VectorXd shape = appr.getShape(Lap<2>());
Eigen::VectorXd expected(9); expected << a, b, b, b, b, 0, 0, 0, 0;
ASSERT_EQ(expected.size(), shape.size());
for (int i = 0; i < expected.size(); ++i) {
EXPECT_NEAR(expected[i], shape[i], 5e-6);
}
double c = 2.*std::exp(3.*(h/s)*(h/s))*h / (-1+std::exp(4*(h/s)*(h/s))) / s / s;
shape = appr.getShape(Der1<2>(1));
expected << 0, c, 0, -c, 0, 0, 0, 0, 0;
ASSERT_EQ(expected.size(), shape.size());
for (int i = 0; i < expected.size(); ++i) {
EXPECT_NEAR(expected[i], shape[i], 5e-8);
}
}
TEST(Approximations, RBFFDGauss2DAugConst) {
double h = 0.1123;
double s = 1.5;
Gaussian<double> g(s);
RBFFD<Gaussian<double>, Vec2d> appr(g, 0);
Range<Vec2d> support = {{0, 0}, {0, h}, {h, 0}, {0, -h}, {-h, 0},
{-h, h}, {-h, -h}, {h, -h}, {h, h}};
appr.compute({0.0, 0.0}, support);
Eigen::VectorXd shape = appr.getShape(Lap<2>());
// computed with Mathematica from (quite long) analytical expression
Eigen::VectorXd expected(9);
expected << -336.1792004798, 88.49955236455, 88.49955236455, 88.49955236455, 88.49955236455,
-4.4547522446, -4.4547522446, -4.4547522446, -4.4547522446;
ASSERT_EQ(expected.size(), shape.size());
for (int i = 0; i < expected.size(); ++i) {
EXPECT_NEAR(expected[i], shape[i], 5e-6);
}
shape = appr.getShape(Der1<2>(1));
// same as without a constant
double c = 2.*std::exp(3.*(h/s)*(h/s))*h / (-1+std::exp(4*(h/s)*(h/s))) / s / s;
expected << 0, c, 0, -c, 0, 0, 0, 0, 0;
ASSERT_EQ(expected.size(), shape.size());
for (int i = 0; i < expected.size(); ++i) {
EXPECT_NEAR(expected[i], shape[i], 5e-8);
}
}
TEST(Approximations, RBFFDPhs) {
Polyharmonic<double, 3> phs;
Monomials<Vec2d> mon(2);
RBFFD<decltype(phs), Vec2d, NoScale, Eigen::PartialPivLU<Eigen::MatrixXd>> appr(phs, mon);
double h = 0.1234;
Range<Vec2d> supp = {{0, 0}, {0, h}, {h, 0}, {0, -h}, {-h, 0},
{-h, h}, {-h, -h}, {h, -h}, {h, h}};
double a = -((356 + 213*std::sqrt(2) + 40*std::sqrt(5) + 275*std::sqrt(10))/1312.0);
Eigen::VectorXd expected(supp.size());
expected << 4 * a - 4, -2 * a + 1, -2 * a + 1, -2 * a + 1, -2 * a + 1, a, a, a, a;
expected /= h * h;
appr.compute({0.0, 0.0}, supp);
auto sh = appr.getShape(Lap<2>());
for (int i = 0; i < supp.size(); ++i) {
EXPECT_NEAR(expected[i], sh[i], 1e-11);
}
/**
* Test computed with Mathematica:
*
* phig = (Exp[-#/sigma^2] &);
* phip = (#^(3/2) &);
*
* $Assumptions = {h > 0, sigma > 0, c > 0, x0 < x, phi[-h] == phi[h],
* phi'[-h] == phi'[h], phi''[-h] == phi''[h]};
*
* pts = {{0, 0}, {0, h}, {h, 0}, {0, -h}, {-h, 0}, {-h,
* h}, {-h, -h}, {h, -h}, {h, h}};
* c = {0, 0};
* phir[x_] := phi[(x - c).(x - c)/h^2];
* n = Length[pts];
*
* A = Table[phir[pts[[i]] - pts[[j]]], {i, 1, n}, {j, 1, n}] //
* Simplify;
* p = {(1 &), (#[[2]] &), (#[[2]]^2 &), (#[[1]] &), (#[[1]] #[[2]] &),
* (#[[1]]^2 &)}; (* , (#^3&), (#^4&)*)
* P = Table[f[(x - c)/h], {x, pts}, {f, p}] ;
* s = Length[p];
* M = Table[0, {i, 1, n + s}, {j, 1, n + s}];
* M[[1 ;; n, 1 ;; n]] = A;
* M[[n + 1 ;; n + s, 1 ;; n]] = Transpose[P];
* M[[1 ;; n, n + 1 ;; n + s]] = P;
* M // MatrixForm
*
* coef = Table[al[i], {i, n}]~Join~Table[be[i], {i, s}];
* rhsphi = Table[
* Laplacian[phir[{x, y} - pts[[i]]], {x, y}], {i, 1, n}] /. {x -> c[[1]], y -> c[[2]]};
* rhsmon = Table[
* Laplacian[f[({x, y} - c)/h], {x, y}], {f, p}] /. {x -> c[[1]], y -> c[[2]]};
* rhs = Join[rhsphi, rhsmon] // Simplify;
*
* Mp = M /. phi -> phip // Simplify;
* rhsp = rhs /. phi -> phip // Simplify;
* MatrixForm /@ {Mp, rhsp}
*
* sol = LinearSolve[Mp, rhsp] // FullSimplify;
* cf = sol[[1 ;; n]];
* uu = Table[u[p], {p, pts}];
* a = -((356 + 213 Sqrt[2] + 40 Sqrt[5] + 275 Sqrt[10])/1312);
* cf2 = 1/h^2 {4 a - 4, -2 a + 1, -2 a + 1, -2 a + 1, -2 a + 1, a, a, a, a};
* cf2 - cf // Simplify
*/
}
TEST(Approximations, RBFFDGauss2DAugOrd1) {
double h = 0.1123;
double s = 1.5;
Gaussian<double> g(s);
RBFFD<Gaussian<double>, Vec2d> appr(g, 1);
Range<Vec2d> support = {{0, 0}, {0, h}, {h, 0}, {0, -h}, {-h, 0},
{-h, h}, {-h, -h}, {h, -h}, {h, h}};
appr.compute({0.0, 0.0}, support);
Eigen::VectorXd shape = appr.getShape(Lap<2>());
// computed with Mathematica from (quite long) analytical expression (same as before)
Eigen::VectorXd expected(9);
expected << -336.1792004798, 88.49955236455, 88.49955236455, 88.49955236455, 88.49955236455,
-4.4547522446, -4.4547522446, -4.4547522446, -4.4547522446;
ASSERT_EQ(expected.size(), shape.size());
for (int i = 0; i < expected.size(); ++i) {
EXPECT_NEAR(expected[i], shape[i], 5e-6);
}
shape = appr.getShape(Der1<2>(1));
// computed with Mathematica from (quite long) analytical expression
expected << 0, 5.94478246428, 0, -5.94478246428, 0, -0.74621135681,
0.74621135681, 0.74621135681, -0.74621135681;
ASSERT_EQ(expected.size(), shape.size());
for (int i = 0; i < expected.size(); ++i) {
EXPECT_NEAR(expected[i], shape[i], 5e-8);
}
}
TEST(Approximations, DISABLED_RBFFDUsageExmaple) {
/// [RBFFD usage example]
double h = 0.1123;
double s = 1.5;
Gaussian<double> g(s);
// Gaussian RBF's augmented with a constant
RBFFD<Gaussian<double>, Vec2d> appr(g, Monomials<Vec2d>(0));
std::cout << appr << std::endl;
// Local neighbourhood
Range<Vec2d> support = {{0, 0}, {0, h}, {h, 0}, {0, -h}, {-h, 0},
{-h, h}, {-h, -h}, {h, -h}, {h, h}};
// compute approximations at point `{0.0, 0.0}`.
appr.compute({0.0, 0.0}, support);
// Get info about the computation.
std::cout << appr.rbf() << std::endl;
std::cout << appr.monomials() << std::endl;
std::cout << appr.center() << std::endl;
std::cout << appr.scale() << std::endl;
std::cout << appr.localCoordinates() << std::endl;
std::cout << appr.getMatrix() << std::endl;
Eigen::PartialPivLU<Eigen::MatrixXd> solver = appr.solver();
// Get shape (stencil weights) for approximation of Laplacian.
Eigen::VectorXd shape = appr.getShape(Lap<2>());
// Get shape (stencil weights) for approximation of value
shape = appr.getShape();
/// [RBFFD usage example]
(void) solver;
}
} // namespace mm